Modeling Action, Knowledge and Control.
Héctor Geffner, Jacques Wainer · 1998
. We formulate a model for reasoning about actions, knowledge and control in the presence of incomplete information and sensing operations. In the model, actions can be non-deterministic, sensors can report the value of either symbols, terms or formulas, and knowledge is constructively defined in terms of observations. The model is related to possible worlds accounts of actions and knowledge but its syntax and semantics are simpler and may be found more useful for the design of actual planners. The model is also related to probabilistic accounts of actions and belief (POMDPs) and can be seen as their logical counterpart. We have used variations of this model for the analysis and synthesis of controllers for various tasks. 1 INTRODUCTION In AI there have been two main approaches to the problem of planning with incomplete information: a procedural approach, focused on extensions of the Strips language and algorithms (e.g., [5, 3]), and a logical approach, focused on extensions of logic ...